Download DOWNLOAD PAPER

DISCUSSION PAPERS IN ECONOMICS
Working Paper No. 10-13
Trade Liberalization, Privatization and Economic Growth
Hang T. Nguyen
University of Colorado at Boulder
November 2010
Department of Economics
University of Colorado at Boulder
Boulder, Colorado 80309
© November 2010 Hang T. Nguyen
1
TRADE LIBERALIZATION, PRIVATIZATION AND ECONOMIC GROWTH
_________________________________________
Hang T. Nguyen
University of Colorado at Boulder
This paper presents a study of the impact of trade liberalization policy on
economic growth with the simultaneous application of privatization policy in 25
transitional countries. The analysis applies two stage least squares to panel data from
1994 to 2006 for these 25 countries. The estimated results provide evidence of a
significantly positive effect of both trade liberalization and privatization on economic
growth, when controlling for political conflict and macroeconomic stability. This chapter
deals with the endogeneity of openness and privatization by using appropriate
instrumental variables and test for the validities of instrumental variables.
1. INTRODUCTION
After the end of the communist era, the Eastern European countries, the
commonwealth of independent states, and countries like Vietnam and China went through
economic reforms that included trade liberalization and privatization, with the hope of
improving their economies to achieve higher growth rates. The simultaneous, rather than
separately timed, enactment of trade liberalization policies and privatization policies has been a
distinguishing characteristic of these transition economies. Therefore, studies that investigate
only the relation between trade liberalization and economic growth, or only the relation
between privatization and economic growth, may not be suitable to analyze transition
economies. Thus, this paper focuses on the simultaneous effects of the two policies on
economic growth.
What are the simultaneous effects of trade liberalization and privatization on economic
growth? Addressing this question is of critical importance to policy makers in transition
countries, who have an interest in finding a reasonable adjustment between the two policies, in
order to improve economic growth.
We have seen many papers observing the separated impact of trade liberalization on
economic growth such as Edwards (1998), Frankel and Romer (1999), Rodriguez and Rodrick
(2000), Chang, Kaltani and Loayza (2005), Romalis (2006) or the separated impact of
privatization on economic growth such as Plane (1997), Cook and Uchida (2003), Godoy and
Stiglitz (2006). But, there are lack of studies on the simultaneous effect of both trade
liberalization and privatization on economic growth. Barlow (2006) was one of the rare papers
that considered the simultaneous effects of both policies, in 22 transition countries from 1993
to 2001. The author found that trade liberalization had a positive effect on economic growth,
while privatization had a negative effect. However, the effect of privatization on economic
growth seems be ambiguous because the coefficient of privatization was sensitive to the data
2
sample and had both negative and positive signs, 1 and more than half of the regressions were
not statistically significant. Barlow also overcame the endogeneity of openness and privatization
by applying Instrumental Variable – Generalized Moment Method (IV-GMM) 2, the instruments
for openness and privatization were the lags, the differences in openness and the differences in
privatization.
Not only Barlow (2006), but many other papers tried to solve endogeneity problems of
openness and privatization. Finding the appropriate instruments was a challenge for economists.
After a long period of examination of this problem, it appeared that there were only two types
of instruments for openness that included the geographical variables recommended by Frankel
and Romer (1999) and the decrease of U.S. MFN tariffs recommended by Romalis (2005). The
instruments of Romalis were probably not suitable for transition countries, and the data were
not sufficient, since the United States was not a large trading partner of many transition
countries. Therefore, the impact of decreasing U.S. MFN tariffs may not have proxied for higher
market access of transition countries in order to push up their trade volumes.
In the course of reviewing the literature, I have found that the important role of
checking whether openness or privatization in their models need to be instrumented, or
whether their instruments are weak or not redundant and exogenous were not recognized. But,
ignoring these tests may lead to biased results.
Thus, there is a gap in the literature on endogeneity that needs to be studied, and that
gap has become an important focus of this research. The contribution of this research project to
literature is three-fold. First, I construct the appropriate instrumental variables for openness and
privatization. Second, I apply econometric techniques to test the endogeneity and the validity of
instruments. Third, I provide a forecast of the simultaneous effects of openness and
privatization on economic growth, the foundation for making macroeconomic policies in the
transition countries.
I use a data set including 25 transition countries during the period from 1994 to 2006.
To overcome the endogeneity of openness and privatization, I apply the two-stage least–
squares test (2SLS) to estimate the effects of openness and privatization on economic growth
under both fixed and random effects. The set of instrumental variables includes the openness
index of trading partners of transition countries, the date of a country’s becoming a WTO
member, the ratio of government debt over GDP, the ratio of public share over private share in
GDP and the product of time and a land-locked dummy. I check the validity of instruments by
applying the Wu-Hausman test, the weak instrument test, the test for redundancy and the test
for overidentifying restrictions. The results from these tests show that openness and
privatization need to be instrumented and that the instruments are valid. I also test fixed effects
versus random effects by using the Hausman test (1978), the Breush-Pagan statistic test, and
1
Barlow cut the original sample into smaller samples and then regressed again. The coefficient signs of
privatization in those small samples were both negative and positive.
2
IV-GMM is method combined instrumental variable method and generalizing method of moment. The
instruments are lag and difference of variable its self. Arellano and Bond (1991) recommend this method.
See chapter 8 in Balgati (2001).
3
the test for fixed effects. The Hausman test (1978) provides evidence for the use of fixed effects.
The primary estimation results suggest that both openness and privatization have a positive
effect on economic growth when political conflict and macroeconomic stability are controlled
for.
I check the robustness of the empirical results in two ways. (1) One way is to classify the
original sample into six groups: Central European, Baltic, European CIS, South East Europe,
Caucasia and Central Asian Countries. Taking out each group from the original sample, and then
running regressions, I have found that the signs of coefficients of all independent variables are
the same as the signs of the coefficients of those variables in the original sample, and they are
mostly significant. (2) The other way is to check the robustness of my estimated results in the
specification of the instrumental variable (the landlocked dummy interacted with time). I
regress the sample in three steps. First, I include the time dummy variable. Second, I replace this
variable by the landlocked dummy variable. Third, I add both the time dummy variable and the
landlocked dummy variable to the instrument set. The statistical results show that coefficients
of all variables in the structural equation are still statistically significant and have the same sign
and small changes in value as the original estimated results.
The rest of this chapter is organized as follows. The second section is a literature review;
the third section describes the methodology; and the fourth section reports the primary
estimation results. The fifth section presents the conclusions.
2. METHODOLOGY
The purpose of this research is to estimate the relations between privatization,
openness and growth, by correcting the endogeneity of openness and privatization. I will use
2SLS estimation of single-equation linear models with the appropriate instrumental variables.
Woodridge (2002) considered the first equation of the structural model (p. 83, eq.5.1):
y1
Y1 γ 1 + X β1 + ε1
=
1
TxK *
Tx ( M −1)
*
(Tx1)
,
(1)
(Tx1)
where y1 is T x 1 vector of the dependent variable; Y1 is T x (M* -1) vector of the endogenous
explanatory variables; X1 is T x K* vector of the exogenous variables included in the 1st equation;
ε1 is T x 1 vector of the error term; T is the number of observations; M* is the number of
endogenous explanatory variables; and K* is the number of exogenous variables included in the
1st equation. In my econometric model, y1 denotes GROW. Y1 includes PRI and OPEN. X1 consists
of WAR, INFLA, INVEST, EMPLOY and PRODU. T = 325, M* = 2 (PRI and OPEN) and K*=5 (WAR,
INFLA, INVEST, EMPLOY and PRODU).
The reduced form of Y1 is (Woodridge, 2002, p. 83, eq. 5.4)
Y1
Tx ( M * −1)
=
X Πˆ
TxK
1
+ Vˆ1
Tx ( M * −1)
,
(2)
4
where X is T x K vector of the determinant variables; Vˆ1 is T x (M*-1) matrix of error term; K is the
number of all determinant variables K=10. X includes X1 and X2. X1 is T x K* vector of the
determinant variables included in the 1st equation. X2 is T x K** vector of the determinant
variables excluded from 1st equation (hereafter called the excluded instruments). K**=K – K* is
the number of determinant variables excluded from equation 1 (K**=5). X2 is assumed to satisfy
the IV exclusion restriction that there is no correlation between X2 and the dependent variable
in the structural equation. In my model, X2 includes instrumental variables: (1) The WTO dummy
variable, (2) The landlocked dummy variable, (3) The openness index of largest trade partners,
(4) The ratio between public output and private output, (5) The ratio between government debt
and GDP. The motivation to choose these instruments is explained in Section 2.4.1, p.35-37.
These excluded instrumental variables are assumed to satisfy the IV exclusion restriction that
they do not correlate with GROW: E (X2, y1) = 0. Arguments about the IV exclusion restriction
should be presented in the following part. The definition of each variable will be given in the
next part.
An estimation result may still be biased if the instruments of endogenous explanatory
variables are not valid. To test the consistency of 2SLS, I have implemented the following tests:
the Wu-Hausman test (See Woodridge, 2002, p.119), the weak instrument test (See Stock and
Yogo, 2004), the redundancy test (See Breusch et al. 1999), and the overidentifying restriction
test -Sargan test (See Woodridge, 2002, p.123).
According to Baltagi (2001), the omitted variables can occur when we cannot include
some necessary variables in the regression model, due to reasons such as unavailability of data
or ignorance. Therefore, the error term will include these omitted variables. If the omitted
variables and the independent variables are correlated, then those variables are endogenous. In
this study, I deal with the edogeneity by applying 2SLS as mentioned above. However, in panel
data, we can see the other occurrence of omitted variables as time-constant variables, called
unobserved effects. Unobserved effects often capture features of individuals (although in my
model, they are features of countries) that do not change over time and they are under two
forms: (1) random effect model and (2) fixed effect model. To deal with the unobserved effects,
I have carried out the following tests: the test for fixed effects (F-test), Breusch-Pagan (1980)
specification test (BP test), Hausman Specification Test (1978) (See Green, 2008, p.194-208).
3. PRIMARY EMPIRICAL RESULTS
3.1 Data
I use a panel data set that covers 25 transition countries from 1994 to 2006. The
transition countries include Albania, Armenia, Azerbaijan, Belarus, Bulgaria, Croatia, the Czech
Republic, Estonia, FYR, Georgia, Hungary, Kazakhstan, Kyrgyz, Latvia, Lithuania, Moldova, Poland,
Romania, Russia, Slovakia, Slovenia, Tajikistan, Turkmenistan, Ukraine and Uzbekistan.
The dependent variable (GROW) is the growth rate of GDP. The main independent
variables are the openness index (OPEN) and privatization (PRI). The level of privatization is
measured by an average of indexes of small-scale and large-scale privatization. The calculation
of indexes of large-scale privatization (LSP) is based on the privatization levels of the assets of
large-scale public enterprises; for example, more than 75%, 50%, or 25% and less than 25%. The
5
calculation of indexes of small-scale privatization (LSP) is also based on the privatization levels of
the assets of small-scale public enterprises. These indexes are established and reported by EBRD.
The openness index is measured by the ratio of trade volumes to GDP. The data on trade
volumes and GDP are also taken from EBRD.
My econometric model includes control variables such as change in investment/GDP
(INVEST), change in employment, change in labor productivity (PRODU), war (WAR) and the
inflation rate (INFLA). 3 The INVEST variable is a proxy for the change in physical capital. The
EMPLOY variable is a proxy for the change in labor. The PRODU variable is a proxy for
technological progress. The WAR variable is a proxy for political conflict, while INFLA is a proxy
for macroeconomic instability. WAR is a dummy variable, which receives the value of one if a
country has political conflict in that year and otherwise is zero. The information related to
political conflict in transition countries in my data set is taken from the website
www.cia.gov/cia/publications/factbook/. The data of all other control variables are obtained
from EBRD.
The instruments for openness and privatization are WTO, PARTNER, TIME_LANDLOCK,
GOVDE and PUB_PRI. WTO is the year in which a country became a WTO member. WTO is a
dummy variable, which receives the value of one if the country began to be or already was a
WTO member in that year and zero otherwise. The data is taken from website www.wto.org.
PARTNER is the average weight of an openness index of the composite of the five largest trading
partners. These openness indexes are also measured by the ratios between trade volumes and
GDP. The information about trading partners and the data on trade volumes of partners are
taken from COMTRADE data set. The data about GDP of trade partners are taken from World
Bank’s data set at the website www.worldbank.org. TIME_LANDLOCK is the multiplication of
two variables: YEAR and LANDLOCK. YEAR is a time variable, which receives value from 1994 to
2006. LANDLOCK is dummy variable, which receives the value of one if a country is landlocked,
and
zero
otherwise.
The
data
on
LANDLOCK
is
taken
from
www.cia.gov/cia/publications/factbook/. GOVDE is government debt over GDP. The data on
government debt is taken from EBRD. PUB_PRI is the ratio of public share of GDP over private
share of GDP. The data on public firms’ share over GDP and private firms’ share over GDP are
obtained from EBRD.
Rose (2004) and Subramanian and Wei (2003) studied the effect of WTO on trade.
According to the authors, when countries became members of WTO, they had to follow bilateral
mutual negotiations on cutting trade protection, and therefore, their openness increased. This
conclusion led to the assumption that having become a WTO member was a important factor in
change in the openness of the country. In this research, therefore, I have constructed a WTO
variable and consider it a potential instrument of openness.
Romalis (2005) showed that a decrease in U.S. MFN tariffs—he assumes that the United
States was the largest trade partner of developing countries—increased the possibility of market
access for a developing country, and therefore increased the openness of that country.
Therefore, the openness index of the largest trade partners of each transition country could
3
The traditional theory of economic growth holds that economic growth is determined by production factors such as
investment, labor and technological progress. Besides, the role of inflation rate and political conflict on the economic
growth of transition countries is affirmed by many authors, such as Barlow (2006).
6
provide necessary information to explain change in the openness index of that country. For this
reason, therefore, I constructed the variable PARTNER and chose it as a potential instrument of
openness.
Frank (1999) used LANDLOCK as an instrument of openness. This arose from the idea
that geographic characteristics of each country provided advantages or created limitations for
that country in trade with the rest of the world. Therefore, I have borrowed from Frank the use
of LANDLOCK as a potential instrument of openness. However, LANDLOCK is a dummy variable
and a time-constant variable, which receives the value of one if the country is landlocked and
zero otherwise. To use it as an instrument for openness in the panel data sample, I multiply the
LANDLOCK variable by the TIME variable to create a new variable, TIME_LANDLOCK. This means
that I consider that LANDLOCK impacts openness through its joint impact with TIME on
openness. This structure translates LANDLOCK into a time series. Section 5 confirms that this
specification of the instrumental variable TIME_LANDLOCK is robust.
During the communist era, the public sector characteristically dominated the whole
economy of a transition country. The operation of public firms did not aim toward profitability,
but only toward social welfare. 4 The government had to use financial sources such as taxes,
foreign debt, and exports to subsidize the operation of public firms. Lacking knowledge of
management, technological progress, the dynamics of competition, and the relationship
between demand and supply, public firms tended to operate ineffectively. The communist
economy suffered from a high pressure of government debt owed to foreign countries. When
the communist government did not have enough capacity to continue subsidizing public firms,
then the government had to sell public firms to domestic or foreign investors. Therefore, higher
levels of government debt created higher motivation for privatization in the transition countries.
Plane (1997) used the ratio of government debt to GDP as the instrument of privatization. In this
paper, I have followed Plane (1997) in choosing the ratio of government debt to GDP as an
instrument of privatization.
Privatization transforms public firms into private firms. The purpose of privatization is
not only to release the government from its heavy financial burden and to improve the
effectiveness of public firms, but also to establish the foundation for market mechanisms
through creating competitive sectors and private firms, and to push up economic growth. On
the other hand, the governments of transition countries think that political institutions need to
be ensured by economic forces such as the contribution of public firms in the economy.
Therefore, the share of public firms’ output to GDP, relative to the share of private firms’ output
to GDP, can be considered important information to explain the change in privatization level. It
follows from this that using only the share of private firms’ output to GDP does not provide
necessary information to explain the change in privatization levels. Therefore, I have adopted
the ratio of public firms’ output to GDP to private firms’ output to GDP as an instrument for
privatization. This is the first use of this ratio in this way.
These instruments are assumed to satisfy the IV excluded restriction that they do not
directly affect growth, but only have direct impact on openness and privatization. To my
knowledge, there are not any theoretical or empirical models that display the direct effect of
WTO, PARTNER, LANDLOCK, and PUB_PRI on economic growth. There is evidence to support the
4
See Matsumara, T. (1998), p.473.
7
relationship between WTO, PARTNER, and LANDLOCK and trade or openness, and none that
shows becoming a WTO member, having sea border, or the openness index of the largest trade
partner has any direct linkage with the economic growth of a country.
There has been some concern about the direct impact of government debt on economic
growth, but few studies have been done concerning this relationship. Recently, Reinhart and
Rogoff (2010) have claimed that debt at low or moderate levels does not affect economic
growth, but that if the ratio between debt and GDP is higher than 90%, then debt does have a
negative effect on economic growth. The claim of Reinhart and Rogoff (2009) has little relevance
to this study. In my sample set, the average level of debt over GDP is 40.5%. There are only 16
observations—the total observations equal 325—that show a ratio between debt and GDP that
is greater than 90%. Examining the case of the Kyrgyz Republic, for example, we find that there
is no linkage between debt and economic growth. The economic growth rate at the high ratio, in
fact, is higher than the economic growth rate at the low ratio. In 1995, the ratio between debts
and GDP was 52.4 % and the economic growth rate was -5.40 %, while in 2000, the ratio
between debts and GDP was 107.34 % and the economic growth rate was 5.44 %. There are also
many cases in which the economic growth rates are very different, although the ratio of debts
over GDP is the same.
The other concern is whether the ratio of public output over private output affects the
economic growth rate. Up to now, there have not been any theoretical or empirical studies that
have mentioned the direct effect of this ratio on economic growth. This ratio is only a
characteristic of transitional economies. The ratio between public output and private output can
be considered an indicator used by governments of transition countries to adjust the
privatization level. I do not find any statistical evidence to support the relationship between
economic growth and the ratio between public output and private output in my sample. Albania
maintained the same ratio between public output and private output (0.333) during the time
period from 1996 to 1999, but the economic growth rate changed within a large range (9.1 % in
1996; -10.2 % in 1997; 12.7 % in 1998; and 10.1 % in 1999). Many other countries provide similar
evidence. For example, from 1994 to 1996, Belarus kept the same ratio between public output
and private output (0.5667), but the economic growth rate changed from -11.7% in 1994 to
2.8 % in 1996. From 1997 to 2000, the ratio between public output and private output was 4,
but the economic growth rate changed (11.4% in 1997; 8.4% in 1998; 3.4% in 1999; 5.8 % in
2000; and 4.7 % in 2001). From 2001 to 2004, the ratio between public output and private
output was 3, but the economic growth rate varied (5.0% in 2002; 7.0 % in 2003; and 11.4 % in
2004).
Table 1 displays a summary of the main variables and also some of the characteristics of
my sample. The observations total 325 and the data on variables such as OPEN, PRI, WAR,
TIME_LANDLOCK, PUB_PRI, WTO, and PARTNER can be fully seen. From 1994 to 2006, the
growth rate of GDP reached its highest rate at 34.5 %. The growth rate of GDP that is higher
than 10 % is about 11% of whole sample. However, 16% of whole sample has negative growth
rate of GDP. The lowest growth rate of GDP is 30.9%. The other feature of the sample is the
inflation rate. 52% of the whole sample has a two-digit inflation rate; 39% of the whole sample
has a three-digit inflation rate; 4% of the whole sample has a four-digit inflation rate. However,
1.8% of the whole sample has a negative inflation rate. The change in investment is also great.
The growth rate of investment is 23.4% on average, 100.49% as the maximum rate and -311% as
the minimum rate. The ratio of public share of GDP to private share of GDP decreases from 1994
8
to 2006. The average ratio is 1.027, the maximum ratio is 5.666, and the minimum ratio is 0.25.
In general, the government debt to GDP reduces. The average level of the government debt to
GDP is 40.5%; the maximum ratio of government debt to GDP is 319.79 % in Hungary in 1994.
Table 1: Statistical Summary of All Variables
Variable
Ob Mean
Std. Dev.
Min
Max
GROW
OPEN
PRI
319
325
325
3.937749
82.95631
3.275462
6.736982
29.42397
0.7658746
-30.9
34.7
1
34.5
241.8
4.165
WAR
INFLA
PRODU
EMPLOY
INVEST
325
318
306
316
290
0.1569231
97.72408
6.450654
-0.2433544
59.61503
0.3642891
458.4971
12.66028
4.310237
190.2608
0
-1.374
-66.2
-22
-31.58657
1
6041.595
84.3
28.3
1965.308
325
325
325
297
325
1040
1.027479
0.4338462
40.50165
69.56384
1000.744
1.092151
0.4963686
32.21594
15.43999
0
.25
0
3.996
27.108
2006
5.666667
1
319.792
115.775
TIME_LANDL~K
PUB_PRI
WTO
GOVDE
PARTNER
Before presenting the estimation results, I provide the results from an econometric test
that show evidence supporting the choice of random effect versus fixed effect model as well as
instruments.
3.2
Random Versus Fixed Effect
The test results are reported in Table 2.
First, I have used the F-test to test the individual effects. The result rejects the null
hypothesis, since F (23, 231) = 5.81 and p = 0.000. Therefore, the unobserved effects and model
must be estimated by either random or fixed effects.
Next, I have used the Breusch-Pagan test to test the null hypothesis H0 that the variance
of unobserved effects (ci) is zero. The test rejects the null hypothesis, since chi-sq (1) = 67.08 and
p = 0.000. This result provides evidence that my model can be estimated either by random or
fixed effects.
The above results did not give concrete direction for my estimate. Therefore, I
employed the test for random versus fixed effects discussed by Hausman (1978). The Hausman
statistic for fixed effects versus random effects reports chi-sq (7) = 54.08 and p = 0.000. Based
on this result, I will estimate my econometric model only by fixed effects.
9
Table 2: Random versus Fixed Effect
Hausman Specific Test
F-Test that all u_i = 0
Breusch-Pagan Test
Chi-sq (7) = 54.08
P-value = 0.000
F(23, 231) = 5.81
P-value = 0.0005
Chi-sq (1) = 67.08
P-value = 0.000
3.3 Testing the Validity of Instruments
The results from testing the validity of instruments are reported in Table 3 with fixed
effects estimation. All tests are passed, to support the validity of all instruments. Table 3 reports
the results of four tests: the endogeneity test, the test for weak instruments, the IV redundant
test, and the Sargan test.
In the fixed effects model, the endogeneity test can be implemented by using the WuHausman test. The Wu-Hausman statistic rejects the null hypothesis that there is no correlation
between GROW and the residuals from the first stage regression, because F (2,229) = 14.89 and
p = 0.000. Therefore, the Wu-Hausman test suggests that both PRI and OPEN are endogenous
variables.
In addition, the joint test of weak instruments provides evidence that the chosen
instruments are not weak. In the first stage regression of PRI, F (5,228) = 42.19 and p = 0.000,
while in the first stage regression of OPEN, F (5,228) = 9.94 and p-value = 0.0000. The Cragg–
Donald Wald F statistic is 8.201. It is higher than 5.91, the critical value for the weak instrument
test of Stock and Yogo (2004), based on 20% maximal IV relative bias and a 5% significance level
for the case of two endogenous explanatory variables and five excluded instruments. It is also
higher than 6.89, the critical value for the weak instrument test of Stock and Yogo (2004), based
on 30% maximal IV relative size and 5% significance level, for the case of two endogenous
explanatory variables and five excluded instruments. Therefore, the instrument set is not weak
in the case of 20% maximal IV relative bias and 30% maximal IV relative size.
10
Table 3: The Statistics of All Tests for the Validity of the Instruments under Fixed Effects
Endogeneity test by Wu-Hausman test
Hypothesis H0: No correlation between GROW and residuals from first stage regression.
If H0 is rejected, there is correlation and PRI & OPEN need to be instrumented.
F (2, 229) = 14.89
P-value = 0.0000
Test of weak instrument
Hypothesis H0 : All instruments are weak
If H0 is rejected, then all instruments are not weak
First stage regression of PRI: Joint test of all instruments
F(5, 228) = 42.19
P-value = 0.0000
First stage regression of OPEN: Joint test of all instruments
F(5, 228) = 9.94
P-value = 0.0000
Cragg-Donald Wald F statistic:
Stock-Yogo weak ID test critical values: 5% maximal IV relative bias
10% maximal IV relative bias
20% maximal IV relative bias
30% maximal IV relative bias
5% maximal IV size
10% maximal IV size
20% maximal IV size
30% maximal IV size
8.201
13.97
8.78
5.91
4.79
19.45
11.22
8.38
6.89
IV redundancy test: (LM test of redundancy of specified instruments)
Hypothesis H0: instrument is redundant or instrument provides no useful information.
If H0 is rejected, then instruments provide useful information and are not redundant.
Instrument
TIME_LANDLOCK
PUB_PRI
WTO
GOVDEB
PARTNER
Chi-sq(2)
9.297
12.217
11.676
7.953
15.911
P-value
0.0096
0.0022
0.0029
0.0188
0.0004
Sargan statistic: (overidentificatio test of all instruments):
Hypothesis H0: all instruments are exogenous
Sargan test statistic: Chi-sq(3) = 5.844, P-value = 0.1174
11
On the other hand, the test for redundancy of each instrument confirms that each
instrument provides useful information to explain PRI and OPEN, hence they cannot be omitted.
Test statistics for TIME_LANDLOCK, PUB_PRI, WTO, GOVDE, and PARTNER are 9.297, 12.217,
11.676, 7.953 and 15.911, respectively. P-values are 0.0096, 0.0022, 0.0029, 0.0188, and 0.0004,
respectively.
Finally, the Sargan test, or the overidentification test, suggests that the
overidentification restriction is valid. The instrumental variables are valid in the sense that they
are uncorrelated with error term in the structural equation. In the fixed effects model, chi-sq (3)
= 5.844 and p-value = 0.1174.
3.4 2SLS Estimated Results
Table 4 reports the estimation result of 2SLS. The column in Table 4 corresponds to 2SLS
estimation’s results for fixed effects. In each cell, the first line reports the estimated regression
coefficient value of the standardized beta coefficient and the level of statistical significance.
Symbols ***, **, * denote 0.01, 0.05, 0.1 significance levels, respectively. The second line
reports standard error. The coefficients of all regressors are statistically significant.
Empirical results from 2SLS estimation under fixed effect suggest that privatization and
openness simultaneously promote economic growth in transition countries. Since both policies
promote the same direction of economic growth, the joint impact of the two policies on
economic growth is automatically predicted as positive.
Table 4: Summary of 2SLS Regression Results
Independent
No standard Beta coefficient
Variable
Dependent variable: GROW
Random
Fixed Effects
effects
PRI
3.088***
6.918***
(0.778)
(1.347)
OPEN
0.100***
0.085**
(0.028)
(0.039)
WAR
-3.546***
-6.090***
(1.069)
(1.521)
INFLA
-0.011***
-0.009***
(0.002)
(0.002)
PRODU
0.194***
0.161***
(0.023)
(0.021)
EMPLOY
0.208***
0.212***
(0.074)
(0.063)
INVEST
0.064***
0.048***
(0.015)
(0.013)
Standard Beta coefficient
Dependent variable: GROW
Random
Fixed Effects
effects
0.351***
0.786***
(0.088)
(0.153)
0.436***
0.371**
(0.122)
(0.168)
-0.191***
-0.329***
(0.057)
(0.082)
-0.786***
-0.644***
(0.127)
(0.134)
0.364***
0.301***
(0.0436)
(0.039)
0.133***
0.136***
(0.047)
(0.040)
0.162***
0.121***
(0.038)
(0.035)
Note: * = 0.1 significant level; ** = 0.05 significant level; and *** = 0.01 significant level
12
My result is different from previous studies: that trade liberalization has a statistically
significant positive effect on economic growth, and that privatization has a statistically
nonsignificant, negative effect on economic growth for transition countries. When comparing
estimation methods, IV-GMM applied by previous studies used only internal instruments,
including first lag and first difference of independent variables, but did not use external
instruments. Internal instruments cannot provide sufficient information to explain endogenous
variables (openness and privatization). In this case, misspecification in application of the
estimation method may have been a reason for the negative sign and statistical insignificance of
the coefficient of the privatization variable.
My findings relate to the impact of production factors on the economic growth of
transition countries, and give evidence to support the traditional theory of endogenous growth.
Those three factors simultaneously contribute positively to economic growth. The standardized
beta coefficients for physical capital, labor and technological progress are 0.121, 0.136, and
0.301, respectively. The standardized beta coefficient of technological progress is highest and is
nearly triple that of physical capital and labor. This shows that technological progress is the most
important factor in production. Transition countries should focus on upgrading advanced
technology to improve their economic growth.
Comparing the standardized beta coefficients in fixed effects estimation, I have realized
that the privatization policy had nearly twice the relative effect on economic growth as the
openness policy, from 1994 to 2006. In fixed effects estimation, the standardized beta
coefficient of privatization is 0.786, while the beta coefficient of openness is 0.371. If openness
were to increase by one standard deviation, economic growth would increase only by 0.371 of a
standard deviation; while if the privatization level were to increase by one standard deviation,
economic growth would increase by 0.786 of a standard deviation. This shows that in order to
accelerate economic growth, governments of transition countries might well emphasize
privatization policy. It also implies that during the period from 1994 to 2006, building the
foundation for market mechanisms through establishing competitive sectors played an
important role in economic reform of transition economies. However, statistical data shows that
openness increased considerably from 1994 through 2006, except for a few countries such as
Albania, Armenia, and Uzbekistan, while the privatization level increased little. The growth rate
of openness reached a maximum of 124.5%, while the growth rate of privatization reached a
maximum of 40%. This may have been caused by the difficulties in implementing the
privatization process, such as asset evaluation of public firms, choosing the privatization method,
procedure in selling stocks and other problems.
In addition, the empirical results show the important role of stabilizing political conflict
and mostly macroeconomic conditions. Political conflict and macroeconomic instability can
decrease considerably the effects of privatization policy and trade liberalization in promoting
economic growth. The standardized beta coefficient of INFLA is -0.644, which is very close to the
value of the standardized beta coefficient of OPEN and PRI under an absolute comparison. To
maintain the positive effects of trade liberalization and privatization policies on economic
growth, governments of transition countries would need to implement macroeconomic
stabilization policies and minimize political conflict. The statistical data from 1994 to 2006
provides promising signs of stabilization of the macroeconomics of transition countries. The
inflation rate of transition countries decreased considerably, from an inflation rate of three or
13
four digits in 1994, to an inflation rate of one or two digits in 2006. In general, political conflict is
well controlled in many transition countries in my data sample, except for Kyrgyz, Georgia,
Moldova, and Uzbekistan.
3.5 Robustness Check
I checked the robustness of the empirical results in two ways. The first was to observe
the changes in the coefficients with different structures of the sample. The second was to check
whether the instrumental variable of the landlocked dummy interacted with time, a reasonable
specification.
First, I classified the original sample into six groups: Central European, Baltic, European
CIS, South East Europe, Caucasia, and Central Asian Countries. I took out each group from the
original sample and then ran the regressions. The statistical results show that signs of
coefficients of all independent variables in the smaller sample are the same as the signs of
coefficients of those variables in the original sample. More than 80% of the total coefficients are
statistically significant. Some loss of significance is expected with the smaller samples. Table 5
reports the results of the robustness check for fixed effect estimation only.
Table 5: Robustness Report of Fixed Effects Estimation by Dropping Countries of One Region
Independent
Variables
PRI
OPEN
WAR
INFLA
PRODU
EMPLOY
INVEST
CONT
(1)
(2)
8.767***
(1.497)
0.047
(0.083)
-5.084
(2.077)
-0.007***
(0.002)
0.156***
(0.025)
0.208***
(0.072)
0.036***
(0.014)
-27.525***
(7.018)
199
7.086***
(1.445)
0.088**
(0.041)
-6.071***
(1.598)
-0.009***
(0.002)
0.163***
(0.022)
0.229***
(0.069)
0.048***
(0.014)
-26.502***
(4.741)
230
Dependent Variable: GROW
(3)
(4)
(5)
6.324***
(1.147)
0.052
(0.036)
-6.682***
(1.884)
-0.008***
(0.001)
0.162***
(0.020)
0.206***
(0.061)
0.035***
(0.012)
-21.758***
(4.313)
221
10.456***
(2.074)
0.042
(0.040)
-4.369**
(1.781)
-0.005**
(0.002)
0.161***
(0.027)
0.364***
(0.079)
0.039**
(0.015)
-34.865***
(6.256)
210
6.748***
(1.408)
0.085**
(0.035)
-6.733***
(1.734)
-0.009***
(0.001)
0.177***
(0.022)
0.194***
(0.067)
0.058***
(0.013)
-26.499***
(4.708)
235
Number
of
Observation
Note:
Central European countries: Czech, Slovak, Hungarian, Poland, Slovenia.
Baltic countries: Estonia, Latvia, Lithuania.
European CIS countries: Belarus, Moldova, Ukraine.
South East Europe countries: Albania, Bulgaria, Croatia, FYR Macedonia, Romania
Caucasia countries: Armenia, Azerbaijan, Georgia.
(6)
3.936***
(1.309)
0.063**
(0.027)
-3.809**
(1.286)
-0.013***
(0.002)
0.141***
(0.019)
0.117*
(0.062)
0.070***
(0.014)
-14.998***
(3.743)
215
14
Central Asian Countries: Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, Uzbekistan
(1) Take Central European Countries out of original sample
(2) Take Baltic Countries out of original sample
(3)Take European CIS countries out of original sample
(4) Take South East Europe countries out of original sample
(5) Take Caucasia countries out of original sample
(6) Take Central Asian Countries out of original sample
* = 0.1 significant level; ** = 0.05 significant level; and *** = 0.01 significant level
Second, I examined the robustness of my estimated results with the instrumental
variable TIME_LANDLOCK. I regressed the sample in three ways. First, I included the time
dummy variables: dummy1994, dummy1995… dummy2006 in the instruments set. DummyT
variable received the value of one for the year of T and zero otherwise (T = 1994, or 1995… or
2006). Second, I replaced the TIME_LANDLOCK variable by the LANDLOCK variable only. Third, I
added both time dummy variables and the LANDLOCK variable in the instrumental set. The
statistical results show that coefficients of all variables in the structural equation are still
statistically significant and have the same sign. The values of coefficients are changed very little
compared with the original estimated results. The P-value of openness is little higher than that
of the original estimated result. All results reported in Table 6 confirm the robustness of my
original estimated results.
Table 6: Robustness Check of the Specification of the TIME_LANDLOCK Variable
Independent
Variable
PRI
OPEN
WAR
INFLA
PRODU
EMPLOY
INVEST
Original
Estimate
Results
6.918***
(1.347)
0.085**
(0.039)
-6.090***
(1.521)
-0.009***
(0.002)
0.161***
(0.021)
0.212***
(0.063)
0.048***
(0.013)
(1)
6.122***
(1.207)
0.054*
(0.031)
-5.861***
(1.391)
-0.009***
(0.001)
0.168***
(0.020)
0.235***
(0.060)
0.048***
(0.012)
Robustness Check
(2)
7.194***
(1.401)
0.073*
(0.043)
-5.806***
(1.568)
-0.009***
(0.002)
0.161***
(0.021)
0.217***
(0.063)
0.048***
(0.013)
(3)
6.122***
(1.207)
0.054*
(0.031)
-5.861***
(1.391)
-0.009***
(0.001)
0.168***
(0.020)
0.235***
(0.060)
0.048***
(0.012)
Note:
(1): Adding time dummy variables
(2): Replace TIME_ LANDLOCK variable by LANDLOCK variable
(3): Adding both time dummy variables and LANDLOCK variable
* = 0.1 significant level; ** = 0.05 significant level; and *** = 0.01 significant level
15
3. CONCLUSION
This paper estimates the simultaneous effect of trade liberalization and privatization
policies on economic growth. I used 2SLS with a full package of tests for endogeneity, weak
instruments, redundancy and overidentifying restrictions. The tests establish the empirical
validity of the model as specified. The main finding is that openness and privatization have
statistically significant and simultaneously positive effects on economic growth. Based on
comparison of beta coefficients, privatization policy effects on economic growth are greater
than trade liberalization policy. In addition, political and macroeconomic stability influence
economic growth considerably. For these countries, a political conflict offsets the contribution
of one standard deviation in privatization and trade liberalization on economic growth. The
robustness check confirms that all empirical results are robust.
References
Baltagi, B., 2001. Econometric Analysis of Panel Data. Chichester, New York: John Wiley.
Barlow, D., 2006. Growth in Transition Economies: A Trade Policy Perspective. Economics of
Transition 14, 505-515.
Bivens, J., & John, I., 2010. Government Debt and Economic Growth. Economic Policy Institute,
Briefing Papers 271.
Breusch, T., Qian, H., Schmidt, P., & Wyhowski, D., 1999. Redundancy of Moment Conditions.
Journal of Econometric 91, 89-111.
Chang, R., Kaltani, L., & Loayza, N., 2005. Openness Can be Good for Growth: The Role of Policy
Complementarities. NBER Working Papers 11787.
Cook, P., & Uchida, Y., 2003. Privatization and Economic Growth in Developing Countries.
Journal of Development Studies 39, 121-154.
Edwards, S., 1998. Openness, Productivity and Growth: “What Do We Really Know? Economic
Journal 108, 383-398.
Frankel, J., & Romer, D., 1999. Does Trade Cause Growth? American Economic Review 89, 379399.
Green, W., 2008. Econometric Analysis. Upper Saddle River, New Jersey: Prentice Hall.
Godoy, S., & Stiglitz, J. E., 2006. Growth, Initial Condition, Law and Speed of Privatization in
Transition Countries: 11 years later. NBER Working Paper 11992.
Matsumura, T. (1998). Partial Privatization in Mixed Duopoly. Journal of Public Economics 70,
473-483.
16
Plane, P., 1997. Privatization and Economic Growth: An Empirical Investigation from a Sample of
Developing Market Economies. Applied Economic 29, 161-178.
Reinhart, C. M., & Rogoff. K. S., 2010. Growth in a Time of Debt. NBER Working Paper 15639.
Rodrick, D., & Rodriguez, F., 1999. Trade Policy and Economic Growth: A Skeptic’s Guide to the
Cross National Evidence. NBER Working Paper 7081.
Romalis, J., 2006. Market Access, Openness and Growth. NBER Working Paper 13048.
Rose, A. K., 2004. Do we Really Know that the WTO Increases Trade? American Economic Review
4, 98-111.
Stock, J. H., & Yogo, M., 2004. Testing for Weak Instrument in Linear IV Regression. NBER
working paper 0284.
Subramanian, A., & Wei, S-J., 2003. The WTO Promotes Trade, Strongly but Unevenly. NBER
working paper 10024.
Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA:
MIT Press.
17